Choosing the best test method for the problem of Heteroscedasticity in a multiple regression model with
 practical application

A thesis Submitted to the council of the college of administration and Economic\ University of Karbala, as partial fulfillment of the requirements for the degree of Master of Science in statistic

By
Raed Asmar Abdullah
Super vised by
Prof. Dr.
Adnan Karim Najim-Aldin

Abstract

This research is considered on attempt to highlight one of the problems of regression, which is one of the basic assumptions underlying ordinary least squares method (OLS). It is considered one of the fundamental conditions  of variance analysis, and in many practical applications this hypothesis cannot achieve thus (OLS) to estimate the model of linear regression is use less (not benefited) to give true and accurate, results and this leads to face the problem of (heteroscedasticity).

 The aim of this thesis is to choose the best parametric test to detect this problem from a group of parametric tests. We wrote an program in (R) language for a comparison in the best of the ratio of the detection true for the presence or absence of a problem through simulation within chapter III has a virtual reality two of the models, the first has a problem of heteroscedasticity the other not contain the problem depending on five tests (Gold field Quant, Prush bejen, White, NCV, Harrison-McCabe).

Chapter four contains the Applied side of the study and depending on real data, in this chapter we choose the best test to detect the problem or not, then we estimate the regression model.

Conclusions and recommendations are found in the end of the study.